A Compact Autonomous Flapping-Wing Aerial Vehicle: Design, Modeling, and Vision-Based Control for Narrow Gap Traversal

Aggressive autonomous flight through narrow gaps presents a critical challenge in drone racing. As flapping-wing aerial vehicles (FWAVs) evolving toward autonomous navigation, avian-inspired FWAVs are significantly hindered by their single-modal flight dynamics and nonlinear underactuated characteristics, leading to the largely unexplored area of narrow gap traversal. To address this gap, this article introduces SparrowHawk, a compact autonomous flapping-wing system optimized for aerodynamic efficiency and structural rigidity. Its design enables a favorable payload-to-weight ratio, accommodating onboard sensing for external-positioning-free localization and maneuverable forward flight. By leveraging the cycle-averaged method for modeling, the system is simplified to a nonlinear time-invariant model, with kinematic and dynamic equations derived in a unified state-space formulation. For robust real-time gate detection during flight, an adaptive Hough transform-based visual algorithm is proposed, yielding pixel-space geometric parameters of gaps without any prior knowledge. The servo control framework employs a closed-loop dual-loop strategy, integrating a TECS-based altitude controller with a cascaded PID attitude controller to ensure precise position and orientation control during autonomous gap traversal. Indoor and outdoor flight experiments demonstrate that SparrowHawk executes tight-radius banked turns and altitude control with a mean error of 0.345 m. The robot successfully traverses an 80 cm-diameter circular gap at speeds up to 5.6 m/s with a 60 cm wingspan, signifying the first achievement for FWAVs. Through multiple repeated trials, it achieved success rates of 80% under normal lighting and 60% under low-light conditions, significantly outperforming a human pilot, the conventional unmodified Hough circle detector, and the deep learning–based YOLOv11 algorithm. This work advances perception-driven control methodologies for FWAVs, paving the way for agile autonomous navigation in confined environments.